Choosing the right sample size is one of the most important steps in planning a survey, experiment, or statistical study. A sample that is too small may produce estimates that are too imprecise or leave a study with insufficient power to detect a meaningful effect. A sample that is unnecessarily large can increase cost, recruitment time, and workload without providing proportional practical benefit.
This Sample Size Calculator helps estimate the minimum number of observations, respondents, or participants needed for several common statistical designs. Depending on the calculation mode, you can account for confidence level, margin of error, population size, response rate, standard deviation, detectable differences, and statistical power.
For survey proportions, the calculator uses a Cochran-style proportion sample-size approach and can apply a finite population correction when the population size is known. It also includes calculations for continuous means, two-proportion A/B tests, reverse margin of error, response-rate adjustment, statistical power visualization, and reference sample-size benchmarks.
Sample-size calculations are design calculations rather than guarantees. The appropriate method depends on the research question, outcome variable, sampling design, statistical test, expected variability or proportion, target effect, significance level, and power. For complex clinical or experimental designs, a dedicated statistical analysis or biostatistical review may be appropriate.
What Is Sample Size?
Sample size is the number of observations included in a study or survey. In a simple survey, this might mean the number of people who provide usable responses. In an experiment, it may mean participants, observations, or measurements allocated to one or more groups.
The required sample size is determined by the statistical objective.
For a proportion survey, the objective may be to estimate a population proportion within a chosen margin of error at a specified confidence level.
For a continuous outcome, the objective may be to estimate a population mean with a chosen precision.
For hypothesis testing or A/B testing, the objective is often to have enough observations to detect a prespecified effect with an acceptable probability of success.
That is why there is no single universal “correct sample size.” Different study designs require different calculations. Sample-size planning generally starts by specifying the statistical question and the inputs relevant to that question.
How This Sample Size Calculator Works
The calculator contains several related sample-size methods rather than forcing every problem into one formula.
Survey and proportion sample size
Use this mode when estimating a population proportion, such as the percentage of respondents with a particular opinion or characteristic. Inputs include confidence level, margin of error, population size when known, and response rate where recruitment planning is needed.
Continuous mean sample size
Use this mode when the outcome is measured numerically and the study is designed around a mean, such as blood pressure, test scores, height, weight, or another continuous measurement.
A/B testing sample size
Use this mode when comparing two conversion rates and planning the required sample per variant at a chosen statistical power.
Reverse margin of error
Instead of asking how many observations are needed for a chosen precision, you can start with a completed sample size and calculate the margin of error achieved under the selected assumptions.
The calculator also provides a statistical power curve and a benchmark matrix so you can see how changing the design inputs affects the required sample.
Survey Sample Size Formula
For a population proportion, a commonly used large-population formula is:
where:
- n₀ = preliminary sample size
- Z = critical value associated with the confidence level
- p = expected population proportion
- E = margin of error expressed as a decimal
For example, a 5% margin of error is:
The sample-size formula shows why precision can become expensive. Because the margin of error is squared in the denominator, reducing the desired error requires a disproportionately larger sample.
Penn State gives the same general relationship for estimating a population proportion and notes that when no reasonable prior estimate of the proportion is available, p = 0.50 is the conservative choice because it produces the largest sample-size requirement.
Why Is p = 0.50 Used?
For the expression:
the maximum value occurs at:
because:
Using 0.50 therefore produces the largest required sample under this simple proportion formula when the true proportion is unknown. That makes it a conservative planning assumption.
It does not mean that every real population has a 50% proportion. If reliable prior information suggests a substantially different proportion, using that information may produce a more efficient sample-size estimate. Penn State's sample-size example explicitly shows that using a known proportion such as 0.25 produces a smaller required sample than the conservative 0.50 assumption.
95% Confidence and 5% Margin of Error: A Classic Example
Consider a survey where:
and no prior estimate of the population proportion is available, so:
For a two-sided 95% confidence level:
The preliminary sample size is:
Because the goal is to obtain a minimum sufficient number of observations, the result is rounded up, giving:
The calculator's audited golden case independently produces 385. Penn State likewise demonstrates that a calculated minimum such as 600.25 must be rounded upward to 601 when determining the minimum sample required.
Does Population Size Affect Required Sample Size?
Yes, but not in a simple one-for-one relationship.
When the population is extremely large compared with the sample, increasing the population further has relatively little impact on the required sample under the basic proportion formula.
When the population itself is finite and not much larger than the required sample, a finite population correction can materially reduce the required sample.
This is why a survey of a population of 1,000 does not necessarily require anything close to 1,000 respondents to achieve a specified precision.
Finite Population Correction
When the population size N is known and finite, the preliminary sample can be adjusted using the finite population correction:
where:
- n₀ = uncorrected sample size
- N = population size
- n = corrected sample size
For example, with a 95% confidence level, 5% margin of error, conservative p = 0.50, and:
the calculator's audited result is:
The implementation applies the FPC to the continuous unrounded preliminary sample and then rounds the final minimum upward. That corrected the previous double-ceiling defect in the calculation engine.
Worked Example: Population of 1,000
Start with:
Then:
Since the required sample must be an integer and must meet or exceed the calculated minimum:
This is why it is important to apply the finite population correction and rounding correctly rather than simply taking an already-rounded 385 and correcting that number.
Confidence Level vs Margin of Error
Confidence level and margin of error answer different questions.
Confidence level controls how much confidence is associated with the interval procedure and therefore determines the relevant critical-value multiplier.
Margin of error describes the desired precision of the estimate.
For a proportion estimate, the general structure is:
and the margin of error depends on a multiplier and the standard error.
Increasing the confidence level generally increases the critical value and therefore increases the required sample size, assuming the other inputs remain fixed.
Reducing the margin of error also increases the required sample size, often substantially.
Why Does a Smaller Margin of Error Require More Participants?
Suppose a survey changes its target from ±5% to ±3%. The investigator is asking for a narrower interval.
Because sample-size formulas generally contain:
in the relevant proportion calculation, reducing E can increase the required sample substantially.
This relationship is not merely theoretical. Penn State's examples show that increasing sample size narrows the resulting confidence interval because standard error decreases as sample size increases.
Sample Size for a Continuous Mean
Some studies are designed around a continuous measurement rather than a proportion. A common normal-approximation planning formula is:
where:
- Z = critical value
- σ = estimated standard deviation
- E = desired absolute margin of error
The relationship is intuitive:
A larger standard deviation means greater uncertainty and therefore a larger required sample.
A smaller desired margin of error requires a larger sample.
A higher confidence level increases the critical value and therefore increases the required sample.
The exact calculation should match the intended study design and assumptions rather than being treated as a universal formula for every continuous outcome.
Worked Continuous-Mean Example
Suppose:
Then:
The minimum integer sample is therefore:
The practical interpretation is that the planning calculation requires at least 97 observations under those assumptions. For an actual study, verify that the assumed standard deviation is appropriate and that the normal approximation matches the planned analysis.
Sample Size and Statistical Power
Sample size is also central to hypothesis testing. Statistical power is commonly written as:
where β is the probability of a Type II error under the specified alternative.
In practical terms, power is the probability of detecting the prespecified effect when that effect is present under the assumptions used in the calculation.
A common planning convention is 80% or 90% power. Increasing the target power generally increases the required sample size because the study is being designed with a lower probability of missing the prespecified effect.
What Does 80% Power Mean?
An 80% power calculation is designed so that, under the specified model and assumed effect, the statistical procedure has about an 80% probability of rejecting the null hypothesis when that effect is truly present.
Equivalently:
The calculation is conditional on its assumptions. It is not a guarantee that an experiment will produce a statistically significant result.
Power depends on factors including effect size, variability, sample size, significance level, and test design. Clinical-study references describe 80% and 90% as common planning choices.
What Happens When You Increase Statistical Power?
Suppose all other inputs stay fixed. Moving from 80% power to 90% power normally requires a larger sample.
The reason is straightforward: the study is being designed to have a greater chance of detecting the prespecified effect if it exists.
The calculator includes an interactive statistical power curve so this relationship can be viewed rather than inferred only from a final sample-size number.
A/B Test Sample Size
A/B testing compares two variants, such as control vs treatment, version A vs version B, or existing conversion rate vs proposed conversion rate.
For a conversion experiment, the key inputs are the expected baseline rate, the expected variant rate, and the desired statistical power. The difference between the rates determines how difficult the effect is to detect.
A small conversion-rate difference requires substantially more observations than a large difference, all else equal.
For two-proportion testing, the exact required sample also depends on the statistical model, significance level, allocation and other assumptions. Different sample-size methods can therefore produce different numerical answers.
Example: 3.0% vs 3.5% Conversion Rate
Suppose:
The absolute difference is:
The relative uplift is:
These are not the same metric. The calculator's audited implementation explicitly distinguishes:
Absolute Difference (MDE): 0.50 percentage points
Relative Uplift: 16.67%
This removed a statistical terminology defect identified during the production audit.
Under the calculator's implemented A/B approximation, the 3.0% vs 3.5% example at 80% power produces:
The audit independently reproduced that value.
Why Small A/B Effects Need Large Samples
A conversion-rate change from 3.0% to 3.5% may look substantial when expressed relatively, but its absolute difference is only 0.5 percentage points. Detecting a small absolute difference against the natural variation in binary outcomes requires a large number of observations.
This is a central reason A/B tests with small conversion lifts can take much longer to reach a reliable sample size.
The general principle extends beyond website testing: a smaller target effect usually requires a larger sample when the desired power and statistical significance remain fixed.
Response Rate and Recruitment Target
A statistical calculation may tell you how many completed responses are needed, but that is not necessarily the number of people you must contact.
If the expected response rate is r, then a simple recruitment adjustment is:
where the response rate is expressed as a decimal.
For example, if 385 completed responses are required and the expected response rate is 80%:
Since you cannot recruit a fraction of a participant:
The calculator's audited response-adjustment case produces exactly 482 invitations. This is a planning adjustment, not a guarantee that 80% of invited people will actually respond.
Reverse Sample Size: What Margin of Error Will My Sample Achieve?
Sometimes the number of participants is already fixed. For example:
Instead of asking: “How many participants do I need for ±5%?” you may want to ask: “What margin of error does a sample of 400 provide?”
Under the calculator's proportion assumptions, the reverse relationship is:
With:
the calculator produces:
For 90% confidence, the same sample produces approximately:
The audited engine independently verifies both reverse-MOE cases.
Sample Size Benchmark Examples
The calculator includes an audited reference matrix to show how confidence and precision affect sample size.
| Population | 95% confidence, ±5% MOE | 95% confidence, ±3% MOE | 99% confidence, ±1% MOE |
|---|---|---|---|
| 100 | 80 | 92 | 100 |
| 500 | 218 | 341 | 486 |
| 1,000 | 278 | 517 | 944 |
| 10,000 | 370 | 965 | 6,240 |
| Large / effectively infinite | 385 | 1,068 | 16,588 |
These values are calculated from the implementation's audited methodology rather than copied from an external sample-size table.
The table also illustrates three important patterns:
A larger confidence level tends to increase sample size.
A smaller margin of error can increase sample size dramatically.
Once the population is sufficiently large, further increases in population size have progressively less influence on the basic survey sample requirement.
Sample Size Is Not Determined by Population Alone
A common misconception is that a population ten times larger automatically requires ten times the sample. That is not generally true for standard proportion estimates.
The required sample is driven by statistical precision and confidence as well as population size. For large populations, the basic infinite-population requirement approaches an asymptotic value under a fixed confidence level, margin of error and proportion assumption.
Finite population correction becomes increasingly relevant when the sample represents a meaningful fraction of the population.
Common Sample Size Mistakes
Using 5 instead of 0.05 for a 5% margin of error
Percent inputs must be converted correctly into decimal probability/precision values before entering the formula.
Rounding in the wrong direction
If the calculated minimum is 384.15, using 384 would leave the sample below the calculated requirement. Minimum sample sizes should be rounded upward.
Forgetting the finite population correction
A known finite population can change the required sample materially.
Applying FPC to an already rounded sample
This can introduce an avoidable rounding distortion. The underlying continuous value should be handled consistently with the selected methodology.
Confusing response rate with completion count
If 385 usable responses are required and only 80% are expected to respond, recruiting 385 people does not provide 385 completed observations.
Confusing relative uplift with percentage-point difference
A move from 3.0% to 3.5% is a 0.5-percentage-point absolute difference but approximately a 16.67% relative uplift.
Treating power as a guarantee
80% power does not guarantee an experiment will detect an effect. It is a design probability under specified assumptions.
Treating one formula as universal
Sample-size methods differ by outcome type and study design. A simple proportion calculation should not automatically be substituted for a power calculation appropriate to a clinical trial or complex experiment.
Survey Sample Size vs A/B Test Sample Size
These calculations look similar because both return a required number of observations, but their statistical purposes are different.
A survey sample-size calculation is often designed to estimate a population quantity with a specified confidence level and precision.
An A/B sample-size calculation is generally designed to detect a prespecified difference between two variants with a chosen significance level and statistical power.
That means you should select the calculation according to the question you are trying to answer, not simply according to which formula produces the smaller sample.
When Should You Use a Sample Size Calculator?
A sample-size calculator is useful before collecting data when you need to plan survey respondents, market-research participants, A/B test traffic, experimental observations, continuous measurements, research participants, prevalence estimates, proportion estimates, or hypothesis-testing studies.
The calculation should be performed before data collection whenever possible because changing the sample-size target after examining results can affect the statistical validity of the study design.
In formal research, document the assumptions used to determine the sample size, including the target outcome, significance level, confidence level where applicable, power, expected variability or proportion, target difference, and any expected attrition or non-response. Sample-size planning is an explicit component of good study design.
Sample Size and Statistical Significance
Sample size and statistical significance are related but are not interchangeable.
A very large sample can make a very small effect statistically detectable. Conversely, a small sample can fail to detect a meaningful effect because statistical uncertainty is large.
Therefore a sample-size calculation should begin with a meaningful target difference or acceptable precision, rather than choosing a large number simply because “more data is always better.”
For clinical research, the target difference should represent the effect that matters for the research question, and the calculation should be based on the planned primary outcome and statistical test.
Sample Size, Margin of Error and Precision
Margin of error is related to the width of a confidence interval. For a confidence interval of the form:
the margin of error represents half of the interval width.
Increasing sample size reduces standard error and therefore generally makes the interval narrower when the other conditions remain comparable.
This is why sample-size planning can be understood as a trade-off:
More observations → greater precision
Fewer observations → less precision
The best design is not necessarily the largest possible sample, but the sample that provides adequate precision or power for the intended question while remaining practical.
How to Use This Calculator
Start by identifying the type of calculation you actually need.
For a survey proportion, enter the confidence level and desired margin of error. Enter the population size when you are sampling from a known finite population. Enter a response rate when you need to convert required completed responses into an estimated recruitment target.
When the sample-size calculation depends on standard deviation and you need to inspect the underlying dispersion of your data first, continue to the Standard Deviation Calculator.
When the calculation is being used to build an interval from a sample estimate, the Confidence Interval Calculator can be used as the next step.
When a standardized effect or normal-theory probability is part of the analysis, the Z-Score Calculator can help with the corresponding standardized-value calculations.
For an A/B conversion experiment, enter the baseline and variant conversion rates and select the desired statistical power.
For reverse MOE, enter the completed sample size and confidence level to estimate the precision achieved.
Review the result together with the displayed assumptions and explanatory text rather than treating the final integer as a universal answer.
How Sample Size Planning Works
Confidence, MOE, Power, Pop
Cochran, t-test, or Casagrande
Integer Ceiling (⌈n⌉)
Divide by Response Rate
Fieldwork Invitations
Planning workflow: Design inputs determine the preliminary formula; the minimum required clean responses are ceiled; non-response rate adjustment calculates the gross invitation count.
Statistical Assumptions Matter
A sample-size result is only as meaningful as the assumptions behind it.
For surveys, consider whether the sampling method is consistent with the calculation. Complex designs such as cluster sampling can require a design effect and therefore a larger sample than simple random sampling. One clinical-research review specifically notes that design effects can increase the required sample when the sampling design differs from simple random sampling.
For experiments, specify the significance level, target effect, power, outcome variability, allocation, and statistical test.
For clinical studies, the primary outcome should drive the sample-size calculation. Expected variability for continuous outcomes and expected proportions for categorical outcomes can materially change the required sample.
What a Sample Size Result Does—and Does Not—Tell You
A sample-size calculation tells you how many observations are required under the specified assumptions and statistical model.
It does not tell you:
- whether the study will ultimately find a significant result
- whether the chosen effect size is scientifically meaningful
- whether participants will respond
- whether the sampling process is unbiased
- whether the population is normally distributed
- whether the planned statistical test is appropriate
- whether missing data will occur
- whether a clinical trial satisfies regulatory requirements
Those are separate methodological questions. A good sample-size calculation is therefore one component of study design, not a substitute for study design.
Frequently Asked Questions
Answers to 28 common questions about sample size determination, margin of error, and statistical power.
Q1What is a good sample size for a survey?
There is no single sample size that is good for every survey. It depends on the desired confidence level, margin of error, expected proportion, population size, sampling design, and other assumptions. For a simple large-population proportion calculation with 95% confidence, ±5% margin of error, and conservative p = 0.50, the familiar result is about 385 completed responses.
Q2How many people do I need for a 95% confidence level and 5% margin of error?
For the standard large-population proportion calculation with p = 0.50, the preliminary result is about 384.15, so the minimum whole-number sample is 385.
Q3Why is the sample size 385?
Because n = [1.96² × (0.5)(0.5)] / 0.05² ≈ 384.15, and the minimum sample is rounded upward to 385.
Q4Is 30 people enough for a survey?
Not as a universal rule. Whether 30 observations are adequate depends on the objective, outcome and required precision or power. A fixed rule such as '30 is enough' cannot replace an appropriate sample-size calculation.
Q5Why is p = 0.5 used in sample-size calculations?
When no reliable estimate of a population proportion is available, 0.50 maximizes p(1 - p), producing the most conservative sample-size requirement for the simple proportion formula.
Q6Does a larger population always require a much larger sample?
No. Once the population becomes large relative to the required sample, the required sample approaches the large-population result under fixed assumptions. Finite population correction matters more when the sample represents a larger fraction of the population.
Q7What is finite population correction?
Finite population correction adjusts a sample-size calculation when the total population N is known and finite: n = n₀ / [1 + (n₀ - 1) / N]. It generally reduces the required sample compared with the uncorrected calculation.
Q8Should I round sample size up or down?
For a minimum required sample, round up, not down. A calculated requirement of 384.15 therefore becomes 385.
Q9What is the margin of error?
The margin of error is half the width of a symmetric confidence interval around an estimate. It represents the desired precision under the stated confidence procedure.
Q10Does a smaller margin of error require more people?
Yes, generally. Because sample-size formulas often contain the inverse square of the margin of error, reducing the target error can require substantially more observations.
Q11Does increasing confidence increase sample size?
Generally yes. A higher confidence level requires a larger critical-value multiplier, which increases the required sample when other inputs remain fixed.
Q12What is statistical power?
Power is 1 - β, where β is the Type II error probability under the specified alternative. It describes the probability of detecting the prespecified effect under the assumptions of the analysis.
Q13Is 80% power enough?
80% is a common planning choice, but whether it is appropriate depends on the study and consequences of a missed effect. Some studies use 90% or another target. Higher power generally requires a larger sample.
Q14What is the difference between confidence level and statistical power?
Confidence level is associated with estimation procedures such as confidence intervals. Power is associated with a hypothesis-testing procedure and the ability to detect a specified effect under the assumed alternative. They address different aspects of statistical design.
Q15How many subjects are needed for 80% power?
There is no universal number. The required sample depends on the statistical test, effect size, variability, significance level, allocation and other design assumptions.
Q16How is A/B test sample size calculated?
It depends on the baseline conversion rate, target variant rate, desired power, significance level, allocation and the statistical approximation used. A smaller detectable difference generally requires more observations.
Q17What is the difference between absolute difference and relative uplift?
Absolute difference is the percentage-point change. For 3.0% to 3.5%, 3.5% - 3.0% = 0.5 percentage points. Relative uplift is [(3.5 - 3.0) / 3.0] × 100 ≈ 16.67%. They should not be used interchangeably.
Q18How many samples do I need for a 3% to 3.5% conversion-rate test?
Under the methodology implemented by this calculator, the audited example produces 19,740 observations per variant at 80% power, or 39,480 total. Different statistical methods can produce different requirements, so the underlying assumptions should be documented.
Q19What happens if the response rate is only 50%?
A simple recruitment adjustment divides the required completed sample by 0.50. For example, 385 completed responses would require an estimated 385 / 0.50 = 770 invitations.
Q20How many people should I invite if I need 385 responses and expect an 80% response rate?
385 / 0.80 = 481.25, so the recruitment target is 482 people.
Q21Can I calculate sample size for a continuous measurement?
Yes, when the calculator's continuous-mean mode is appropriate. A common normal-approximation form is n = (Zσ / E)², where σ represents the estimated standard deviation and E the desired absolute precision.
Q22What if my population size is unknown?
For the simple proportion calculation, leaving population size unspecified means the calculator treats the population as effectively infinite and does not apply finite population correction.
Q23What if my sample size is already fixed?
Use the reverse margin-of-error calculation to estimate the precision achieved under the calculator's selected assumptions.
Q24Does sample size guarantee statistical significance?
No. Sample size is planned using assumptions about effect size, variability, significance level and power. It cannot guarantee the result of a future analysis.
Q25Can I use this calculator for clinical trials?
It can provide calculations for common simplified designs, but clinical trials often require design-specific methods involving the primary outcome, effect size, variance, allocation, significance level, power, dropout, interim analyses and other considerations. More complex trials should be reviewed using an appropriate statistical method and, where necessary, by a biostatistician.
Q26Does this calculator account for every sampling design?
No. Simple sample-size formulas do not automatically account for every complex sampling design. Clustered or otherwise complex survey designs can require additional adjustments such as a design effect.
Q27Why can two sample-size calculators give different answers?
Different calculators may use different statistical models, critical values, continuity corrections, allocation assumptions, prevalence assumptions, rounding conventions, or power approximations. The formula and assumptions should therefore be checked before comparing numerical results.
Q28Should I always choose the largest sample-size estimate?
No. The appropriate calculation is determined by the design and statistical objective. An unnecessarily large sample can increase cost and recruitment burden, while an insufficient sample can reduce precision or power.
Final Takeaway
Sample size is a design decision built around a statistical objective.
For survey estimation, the major drivers are confidence level, margin of error, population size and the expected proportion.
For continuous outcomes, variability and desired precision become central.
For A/B testing and hypothesis testing, effect size, significance level and statistical power are essential.
For recruitment planning, response rate or attrition may require an additional adjustment.
The most useful sample-size calculation is therefore not simply the one that produces a number. It is the one whose formula, assumptions, inputs and interpretation match the question being studied.
This calculator is designed to make those relationships visible: it provides the resulting sample size, supporting calculations, power visualization, benchmark comparisons and reverse margin-of-error analysis so the number can be understood rather than treated as a black-box recommendation.
This calculator provides statistical estimates based on the selected calculation method and supplied assumptions. Results should be interpreted as planning values, not universal guarantees.
For surveys, confirm that the sampling design and assumptions are appropriate for the intended population. For experiments and A/B tests, verify the statistical test, significance level, target difference, allocation and power assumptions. For clinical or regulated research, use the method required by the study protocol and applicable methodological guidance.
A sample-size calculation does not guarantee response rate, representativeness, statistical significance, or scientific importance of an observed result.